Detailled semantic labeling (3D) result


Name Ramiya, A. M. et al. (orthoimage used in addition)
Affiliation Indian Institute of Space Science and Technology, India
Abbreviation IIS_1
Strategy (u)nsupervised, (s)upervised, (h)ybrid s

=============================================================
↓ predicted || reference →
 power
low_veg
imp_surf
car
fence_hedge
roof
fac
shrub
tree
missing points
power
---
0.000
0.000
0.035
0.027
0.225
---
0.002
0.002
426
low_veg
---
0.524
0.252
0.007
0.038
0.077
---
0.058
0.023
2179
imp_surf
---
0.053
0.934
0.001
0.002
0.005
---
0.000
0.000
342
car
---
0.000
0.056
0.696
0.114
0.047
---
0.004
0.002
301
fence_hedge
---
0.003
0.040
0.012
0.447
0.266
---
0.097
0.116
148
roof
---
0.000
0.001
0.003
0.011
0.954
---
0.014
0.012
500
fac
---
0.022
0.017
0.026
0.093
0.590
---
0.039
0.060
1711
shrub
---
0.033
0.034
0.008
0.227
0.331
---
0.244
0.097
596
tree
---
0.003
0.007
0.001
0.157
0.240
---
0.202
0.361
1495
Precision/Correctness
 --
0.884
0.780
0.593
0.137
0.731
--
0.238
0.724
Recall/Completeness
 --
0.524
0.934
0.696
0.447
0.954
--
0.244
0.361
F1
 --
0.658
0.850
0.640
0.210
0.828
--
0.241
0.482

Overall accuracy 0.699

Data Download


Download a zip file containing laz files per class. Those can be visualized with CloudCompare

Click here for download!

Data previews (derived from voxelisation at 50cm voxelcube)

Participant labeling (lowest points when multiple labels at one X/Y position)

Red/green image, indicating wrongly classified areas

!!! All red/green visualisations (also the point clouds in the zip file) are defined from the reference point of view.